Tem Images of Membranes: a Multiresolution Edge-detection Approach for Watershed Segmentation

This paper presents a method for the segmentation of membrane images acquired with transmission electron microscopes (TEM) for classification purposes. It relies on a multiresolution representation of the image to detect the edges at appropriate scales by an adaptive thresholding of the gradient images. The edges found at the different scales are combined and homogeneous regions are extracted with the watershed algorithm. The method reduces the tendency of the watershed algorithm to over-segment and results in closed homogeneous regions. The tests done on several types of images acquired with several microscopes reveal a promising segmentation method. It works well in the majority of cases and is appropriate to build up the desired classification.